We examine how principals should design incentives to induce time-inconsistent procrastinating agents to complete tasks efficiently. Delay is costly to the principal, but the agent faces stochastic costs of completing the task, and efficiency requires waiting when costs are high. If the principal knows the task-cost distribution, she can always achieve first-best efficiency. If the agent has private information, the principal can induce first-best efficiency for time-consistent agents, but often cannot for procrastinators. We show that second-best optimal incentives for procrastinators typically involve an increasing punishment for delay as time passes.